single source · data curation
Adaptive RL prompt and environment curriculum
Adaptive RL prompt and environment curriculum sorts prompts into three buckets based on their estimated pass rate: hard, solved, and trainable.1 It focuses on prompts whose groups carry learning signal and scales down the weight of environments that run out of such prompts.2
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
The curriculum sorts the prompts of each environment into three buckets by their estimated pass rate, namely hard prompts with an estimate of 0, solved prompts with an estimate of 1, and trainable prompts in between
— tech reportThe curriculum therefore focuses on prompts whose groups carry learning signal, and across environments, it scales down the weight of an environment that runs out of such prompts
— tech report
Also called The curriculum.
Filed under data curation › data mixture & curriculum